DualGCN: Dual-Branch Graph Distillation for Cold-Start Recommendation
What it is
This paper proposes DualGCN, a framework that uses dual-branch graph knowledge distillation to solve the cold-start problem in recommendation systems. It decouples training into a teacher network that anchors item representations with semantic priors and a student branch that aligns these with collaborative signals using a magnitude-agnostic cosine objective.
The problem it solves
Graph Collaborative Filtering models suffer from catastrophic representation collapse for new items because they are topologically isolated, and directly adding semantic features often causes optimization conflicts and representation degradation.
What you could use it for
- Improving recommendation accuracy for new items with no interaction history
- Integrating multi-modal semantic features into graph-based collaborative filtering without degrading performance for established items
- Resolving representation collapse in isolated nodes within graph neural networks
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.